Diagonal rod length issue

Achieved calibration of 0.1 mm accuracy using Rich.s Marlin
Current Delta geometry values:
X (Endstop Adj): -1.575
Y (Endstop Adj): -1.175
Z (Endstop Adj): -2.987
A (Tower A Position Correction): 0.00
B (Tower B Position Correction): 0.00
C (Tower C Position Correction): 0.00
I (Tower A Radius Correction): -2.75
J (Tower B Radius Correction): -4.00
K (Tower C Radius Correction): -2.00
R (Delta Radius): 283.21
D (Diagonal Rod Length):448.95
H (Z-Height): 315.1

My initial parameters were:
X (Endstop Adj): 0
Y (Endstop Adj): 0
Z (Endstop Adj): 0
A (Tower A Position Correction): 0.00
B (Tower B Position Correction): 0.00
C (Tower C Position Correction): 0.00
I (Tower A Radius Correction): 0
J (Tower B Radius Correction): 0
K (Tower C Radius Correction): 0
R (Delta Radius): 275.159
D (Diagonal Rod Length):440.5
H (Z-Height): 318.5

Took 13 iterations to calibrate at 0.1 accuracy.
The Diagonal rod length and delta radius are very accurately measured but the firmware changes them while calibration.
Surprisingly, the XY dimensions are accurate at diagonal rod length of 448.95 mm and not at the measured value of 440.5 mm
So what could be the reason behind it ?
Also changing the IJK values to -0.75, -2, 0 (that is minus the lowest value) destroys the calibration completely.
And how shall i proceed further to achieve accuracy upto 0.05mm ?

I don't know about Rich's Marlin ABL, but I would start with reducing the endstop offsets ( keep them within 1mm difference or better ) and adjust z-height.
If possible, keep the rod length fixed, that keeps the number of iterations lower.

How are you measuring your arms? It needs to be measured from the pivot point to pivot point.

Also, dimensional scale is not only dependent on arm length, but also horizontal radius and endstop adjustments. It's entirely possible to have roughly accurate scaling with not-real-world numbers, or to have a condition where the parts are not evenly scaled, but they still come out close enough that they seem to be the right size. Often, printing a much larger object will reveal more discrepancy.

Edited 1 time(s). Last edit at 08/27/2016 11:58AM by n8bot.